Dasari et al. (2025) A regionalization based machine learning framework for bias correction and downscaling of ESACCI soil moisture in data limited region: A case study over India

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-11-25
• Authors: Indhu Dasari, Vamsi Krishna Vema
• DOI: 10.1016/j.jhydrol.2025.134657

Research Groups

• Department of Civil Engineering, National Institute of Technology Warangal, Warangal, India

Short Summary

This study developed a regionalization-based machine learning framework for bias correction and downscaling of ESACCI soil moisture data in data-limited regions like India, demonstrating significant bias reduction (over 90%) and effective downscaling with high containment ratios (over 89%).

Objective

• To develop and test a regionalization-based machine learning framework for bias correction and downscaling of European Space Agency Climate Change Initiative (ESACCI) soil moisture products in data-limited regions, using India as a case study.

Study Configuration

• Spatial Scale: India, with soil moisture data processed at resolutions of approximately 83.25 km x 83.25 km (0.75° x 0.75°), 27.75 km x 27.75 km (0.25° x 0.25°), and 9.21 km x 9.21 km (0.083° x 0.083°).
• Temporal Scale: Not explicitly stated in the provided text.

Methodology and Data

• Models used: Neural forests with leave-one-out method (for bias correction), Artificial Neural Network (ANN) model (for downscaling).
• Data sources: European Space Agency Climate Change Initiative (ESACCI) soil moisture product, in-situ soil moisture station data across India.

Main Results

• Initial assessment revealed non-uniform relative biases in ESACCI soil moisture across Indian stations, ranging from 0.07 to 3.98.
• The proposed bias correction method using neural forests substantially reduced biases, achieving reductions exceeding 90% in several locations.
• Downscaling of bias-corrected soil moisture from approximately 83.25 km x 83.25 km to 27.75 km x 27.75 km, and subsequently to 9.21 km x 9.21 km, was successfully performed using an ANN model.
• Validation of the downscaled data against original and in-situ data showed containment ratios exceeding 89% for all stations, indicating high efficacy of the downscaling framework.
• The developed frameworks are effective for bias correction and downscaling of soil moisture data in data-limited regions.

Contributions

• Introduction of a novel regionalization-based machine learning integrated framework for both bias correction and downscaling of satellite soil moisture data.
• Demonstrated high efficacy of the framework in significantly reducing biases (over 90%) and accurately downscaling soil moisture in data-limited regions, using India as a comprehensive case study.
• Provides a robust methodology for improving the utility of satellite soil moisture products for hydrological applications in regions with scarce in-situ observations.

Funding

• Not explicitly stated in the provided text.

Citation

@article{Dasari2025regionalization,
  author = {Dasari, Indhu and Vema, Vamsi Krishna},
  title = {A regionalization based machine learning framework for bias correction and downscaling of ESACCI soil moisture in data limited region: A case study over India},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134657},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134657}
}